JP2001099710A - Spectrum estimating method of spectral reflectance of multi-band image and spectrum estimating system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a spectrum estimating method of a spectral reflectance of a multi-band image and a spectrum estimating system, capable of shortening a processing time for obtaining an estimated spectrum without reducing estimation accuracy of the spectrum effectively, when the multi-band image is photographed by using a variable wavelength filter and the spectrum of the spectral reflectance of a photographing object is estimated by using it. SOLUTION: In this estimating method, a conversion table in which a brightness value obtained by photographing a chart having a known reflectance is made to correspond to the reflectance in each channel of a multi-band image comprising plural original images, is formed beforehand, and the brightness value of the original images of the multi-band image obtained by photographing an object is converted into a reflectance by using the conversion table, to thereby estimate a spectrum of a spectral reflectance of the object. This spectrum estimating system of the spectral reflectance of the multi-band image by using the estimating method of the spectrum is also provided.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a spectral multi-band image for estimating a spectral reflectance spectrum of a photographed object from a multi-band image photographed on a plurality of channels having different wavelength ranges using a wavelength variable filter. The present invention relates to a reflectance estimation method and a spectrum estimation system using the same.

[0002]

2. Description of the Related Art Today, with the advance of digital image processing, a wavelength region is different besides a color image formed by three images using three primary colors of red (R), green (G) and blue (B). A multi-band image obtained by photographing the same subject on at least four channels or more is used. For example, in the field of arts and crafts such as paintings, in order to faithfully reproduce colors, the above multi-band images are used for organizing and saving. Also, when developing a film capable of obtaining an image with good color reproduction or an image sensor of a digital camera, a multiband image must be obtained by multiplying the spectral reflectance of the subject by the spectral sensitivity to obtain accurate image data. Is used to estimate the spectral reflectance spectrum. In particular, since the spectral sensitivity of a film and the spectral sensitivity of an image sensor of a digital camera are designed using the spectral reflectance of a subject, it is important to accurately estimate the spectral reflectance of the subject.

[0003] In order to estimate the spectral reflectance of a photographic subject from such a multi-band image, a multi-band image spectrum estimating system for estimating the spectrum of the multi-band image is generally used. This multi-band image spectrum estimating system generally includes a wavelength tunable filter that transmits reflected light of a subject corresponding to at least four or more channels having different wavelength ranges when capturing a subject, CCD that receives light and acquires the original image as a digital image for each channel
A camera, a software having a processing algorithm for estimating a spectral reflectance spectrum of a subject from an original image acquired by the CCD camera, and a spectral reflectance spectrum estimating apparatus including a computer executing the software. .

Here, the above-mentioned spectral reflectance spectrum estimating apparatus generally estimates the spectral reflectance spectrum by the following method. That is, the output signal value V _i (i = 1 to 1) which is a measurement value obtained by receiving the original image by the CCD camera.
n, n are the number of channels), that is, the QL value which is the image data value of each channel of the multi-band image is obtained by using a known illumination light spectral intensity distribution E (λ) at the time of shooting and a known CC
The spectral sensitivity distribution S (λ) of the D camera, the known spectral transmittance distribution f _i (λ) of the wavelength tunable filter, and the known spectral transmittance distribution L (λ) of the optical system are represented by Expression (1). The spectral distribution O (λ _j ) of the spectral reflectance specific to the subject (j = 1 to m, where m is the subject) is obtained by decomposing the integral of Expression (1) and deriving the matrix represented by Expression (2). (The number of measurement points of the spectral reflectance). When the number n of channels is equal to the number m of measurement points of the spectral reflectance of the subject, the matrix F in Expression (2) becomes a square matrix, and the inverse matrix of the matrix F can be analytically obtained. Since the number of measurement points of the spectral reflectance m is smaller than the number m, the estimation is performed by the known Wiener estimation method (by the least square method) using the estimation matrix G corresponding to the pseudo inverse matrix of the matrix F as shown in Expression (3). . Here, E _k L _k S _k · f _l is E (λ _k ) · L
(Λ _k ) · S (λ _k ) · f _l (λ _k ) · Δλ, and O
_k is O (λ _k ).

(Equation 1)

(Equation 2)

(Equation 3)

In the spectral reflectance spectrum estimating apparatus, in addition to the above estimating method, an eigenvector composed of a spectral waveform specific to a photographing subject is obtained in advance, and the spectral distribution of the spectral reflectance is estimated using the eigenvector. Principal component analysis, such as performing the analysis, is also used.

[0006]

However, in the former estimation method, when the number n of channels is equal to the number m of measurement points of the spectral reflectance of the object, the spectral distribution of the spectral reflectance is analytically obtained as described above. Although it is possible, in order to perform more accurate spectral reflectance estimation, the number m of measurement points of the spectral reflectance of the subject and the number n of channels are increased.
The size of the matrix must be increased, so that it takes time to find the inverse of the matrix F in equation (2), and O
(Λ _k ) cannot be obtained in a relatively short time. Also,
Also in the Wiener estimation method, the estimation matrix G corresponding to the pseudo inverse matrix of the matrix F becomes an m × n matrix, the size of the matrix increases, and it takes time to calculate the estimation matrix G. As a result, there is a disadvantage that it takes a lot of time to estimate the spectrum of the spectral reflectance of the subject. In the latter estimation method based on principal component analysis, if the shooting subject is known, the spectral distribution of the spectral reflectance can be accurately estimated using the eigenvector, but the shooting subject is not fixed and the eigenvector is unknown. In this case, there is a problem that the accuracy of the spectral distribution of the estimated spectral reflectance is deteriorated.

[0007] Originally, the spectrum of the spectral reflectance is
It is represented by a smooth curve, and the spectrum obtained by the above method is represented by the spectral reflectance of the wavelength region for each channel. Although desirable, a smooth spectrum could not be obtained due to linear interpolation. Furthermore, the spectral transmittance of a liquid crystal tunable filter used as a wavelength tunable filter has a temperature dependence in which the peak wavelength and the half-value width change with the temperature of the liquid crystal. In the conventional spectral reflectance spectrum estimating method for estimating the spectrum, only an incorrect spectral reflectance spectrum was obtained.

As described above, when estimating the spectrum of the spectral reflectance, this processing takes a long time to calculate an inverse matrix having a large matrix size, and the processing method is complicated. In addition, the Wiener estimation method by the least squares method used when the number of measurement points m of the subject spectral reflectance is larger than the number of channels n minimizes a noise component included in an original image constituting a captured multiband image. Rather than being used to suppress by the square method, it is used only to increase the number of estimated points of the spectrum of the spectral reflectance to be estimated.Moreover, the matrix size when obtaining the pseudo-inverse matrix increases, and a large processing time is taken. Needed. In the method based on the principal component analysis, as described above, in the case of a photographing subject that is not represented by the eigen spectrum prepared in advance, the accuracy of spectrum estimation is poor. Furthermore, when the number of wavelengths of the spectrum estimated according to the spectrum of the spectral reflectance, which is a smooth curve, is increased by linear interpolation, only a spectral distribution having irregularities different from the original curve was obtained. Furthermore, when a liquid crystal tunable filter is used as a wavelength tunable filter, the spectrum of the estimated spectral reflectance is inaccurate because the peak wavelength and the half-value width of the spectral transmittance of the liquid crystal tunable filter are temperature-dependent. Was.

In order to solve the above-mentioned problem, the present invention takes a multi-band image using a wavelength tunable filter and estimates the spectral reflectance spectrum of the photographed subject using the image. It is an object of the present invention to provide a spectrum estimation method and a spectrum estimation system of a spectral reflectance of a multiband image, which can reduce a processing time for obtaining an estimated spectrum without lowering a spectrum estimation accuracy.

[0010]

In order to achieve the above object, the present invention divides a photographing wavelength region into a plurality of channels by using a wavelength variable filter, and photographs the same subject corresponding to each channel. Obtaining a multi-band image composed of a plurality of original images photographed by the method, and estimating the spectrum of the spectral reflectance of the subject from the multi-band image, wherein the wavelength variable filter has one channel.
A wavelength tunable filter having 0 or more channels and an average half-width of the spectral transmittance distribution of all channels of 40 nm or less, and a QL value obtained by photographing a chart having a known spectral reflectance for each channel. Is converted in advance into a conversion table in correspondence with the spectral reflectance, and the QL value of the original image of each channel of the multi-band image is converted into the spectral reflectance using the conversion table, thereby obtaining the spectral reflectance of the subject. The present invention provides a method for estimating a spectral reflectance spectrum of a multiband image, which is characterized by estimating a spectrum of a multiband image. Here, the variable wavelength filter refers to a filter that electrically changes the spectral transmittance instantaneously with one element.

In this case, it is preferable that the tunable filter is a liquid crystal tunable filter, and the tunable filter in the wavelength region of the channel corresponds to the estimated spectral distribution of the spectral reflectance of the subject. Using one device function determined by the product of the spectral transmittance distribution of the filter, the spectral sensitivity distribution of the photographing means, and the spectral intensity distribution of the photographing illumination light at the time of photographing, the spectral reflection of the subject is performed by performing deconvolution processing. Preferably, the rate spectrum is calibrated. Here, the liquid crystal tunable filter has a multi-stage structure in which a birefringent crystal element and a liquid crystal element are sandwiched by a deflecting plate. This is a liquid crystal filter (Lyot type) in which the center wavelength of transmittance can be arbitrarily selected. At this time, the deconvolution processing is preferably performed using a matrix operation. When the number of pixels is small, for example, when the number of pixels is less than 100,000 pixels, the deconvolution processing is preferably performed using a Fourier transform, and the Fourier transform is used. In this case, it is preferable to use FFT. In addition, it is preferable that a device function having a value within a range of 50 nm above and below the center wavelength of a wavelength region captured by the channel is selected as the device function.

Further, before performing the deconvolution processing, the wavelength of the spectrum of the spectral reflectance of the subject is compressed or expanded according to the half-value width of the variable wavelength filter, and the deconvolution processing is performed. It is preferable to perform an inverse transformation to expand or compress the wavelength of the spectrum of the spectral reflectance of the subject obtained by the deconvolution processing, corresponding to the compression or expansion of the wavelength. Further, when performing the deconvolution processing, it is preferable to perform filter temperature correction for compensating for the temperature dependence of the wavelength tunable filter, and the filter temperature correction has a known spectral reflectance distribution and is abrupt with respect to wavelength. A color chart with high saturation indicating a change is photographed, and a spectral reflectance distribution of the color chart is obtained using the captured color chart, and a wavelength at which a waveform of the obtained spectral reflectance distribution matches a known waveform of the spectral reflectance. It is preferable that the shift amount of the object is calculated, and the spectral reflectance distribution of the subject is corrected in accordance with the obtained shift amount. Further, the tunable filter includes:
A liquid crystal tunable filter including a liquid crystal unit, wherein the filter temperature correction includes a temperature of the liquid crystal unit obtained from a thermometer in contact with the liquid crystal unit of the liquid crystal tunable filter and a spectral transmittance distribution of the liquid crystal tunable filter. Is stored in advance as characteristic data, and the shift amount of the wavelength of the spectral transmittance distribution of the liquid crystal tunable filter is obtained from the temperature of the liquid crystal unit at the time of capturing a multi-band image,
The spectral reflectance distribution of the subject may be corrected according to the shift amount.

Further, before performing the deconvolution processing, the number of spectral reflectance spectrum data is increased by nonlinear interpolation between the center wavelengths of the respective channels with respect to the spectral reflectance spectrum of the subject. It is preferable to perform the deconvolution processing using the spectrum of the increased spectral reflectance. It is preferable that the spectrum of the spectral reflectance with the increased number of data has a wavelength interval of 10 nm or less.

According to the present invention, there is provided a photographing means for photographing a subject, and when photographing the subject with the photographing means, a photographing wavelength region is divided into a plurality of channels, and reflected light of the subject is reflected in a wavelength region corresponding to the channel. And a conversion table in which a QL value obtained by photographing a chart having a known spectral reflectance is associated with the spectral reflectance for each of the channels, and the imaging means and the wavelength Spectral reflectance for estimating the spectrum of the subject's spectral reflectance by converting the QL value of the original image of each channel of the multi-band image captured using the variable filter into the spectral reflectance using the conversion table A spectrum estimation system for a multi-band image, comprising: a spectrum estimation device.

At this time, the spectral reflectance spectrum estimating apparatus includes a spectral transmittance distribution of the wavelength tunable filter transmitting the reflected light of the subject in the wavelength region of the channel, a spectral sensitivity distribution of the photographing means, Using one device function determined by the product of the spectral intensity distribution of the photographing illumination light and the estimated spectral reflectance spectrum of the subject, deconvolution processing is performed to calibrate the spectral reflectance spectrum of the subject. Is preferred.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a method and system for estimating the spectral reflectance of a multiband image according to the present invention will be described in detail with reference to the preferred embodiments shown in the accompanying drawings.

FIG. 1 illustrates a method for estimating the spectral reflectance of a multi-band image according to the present invention, and is also an example of a system for estimating the spectral reflectance of a multi-band image according to the present invention. Of the spectrum estimation system (hereinafter, referred to as the present system) 10 of FIG. The system 10 includes a CCD camera 12 as a photographing means for photographing the same subject in at least 10 channels having different wavelength regions, and a subject reflected light in a wavelength region corresponding to a channel when the subject is photographed by the CCD camera 12. Liquid crystal tunable filter 14 as a wavelength tunable filter that transmits light, and a spectral reflectance spectrum estimating device 16 for estimating the spectral reflectance spectrum of the photographing subject
And is configured.

The CCD camera 12 has a plurality of channels,
For example, a CCD device in which transmitted light obtained through a liquid crystal tunable filter 14 whose spectral transmittance distribution is controlled so as to transmit light in a specific wavelength region for each channel of 16 channels, which is described later, is arranged in an area. Is an imaging means for forming an image on the light receiving surface and obtaining an original image of an imaging subject for each channel having a different wavelength region. That is,
If the number of channels is 16, 16 original images are obtained. The number of pixels and the pixel size of the CCD camera 12 are not limited, and are known sizes, for example, 1024 × 102
4 pixels, a known pixel size, for example, 12 × 1
It may be 2 μm.

The liquid crystal tunable filter 14 controls the control voltage so as to transmit the reflected light from the photographing subject in a specific wavelength region for each of a plurality of channels, thereby providing a desired band-pass filter characteristic. A wavelength tunable filter to be obtained, wherein the number of channels is 10 or more, and the average wavelength half width of the spectral transmittance distribution of all channels is 40 nm or less. The upper limit of the number of channels is preferably about 80, and the lower limit of the half width is preferably about 5 nm. The half width is more preferably 15 to 35 nm. The liquid crystal tunable filter 14 can obtain, for example, a band-pass filter having a peak value as shown in FIG. 2 corresponding to the number of channels set by the operator and the wavelength region corresponding to each channel. FIG. 2 shows an example in which the imaging wavelength region of 410 to 710 nm is divided by 16 channels. As described above, the reflected light of the subject transmitted through the band-pass filters provided for each channel by the liquid crystal tunable filter 14 is received by the CCD camera 12 as transmitted light of a specific wavelength, and an output signal is obtained. The number of channels of a conventional filter for obtaining a multiband image is 6 to 8 channels. However, in the liquid crystal tunable filter 14, the number of channels can be 12 or more. In this embodiment, the liquid crystal tunable filter 14 is used as a wavelength tunable filter. However, any filter that can electrically change the spectral transmittance instantaneously with one element may be used. Alternatively, an acousto-optic filter may be used.

The spectral reflectance spectrum estimating device 16 calculates C
A data processing unit that A / D converts an output signal obtained by receiving light from the CD camera 12 to obtain digital image data, corrects a DC offset, and converts the output signal into image data of an original image forming a multiband image. 18 and the relationship between the image data value of the gray patch, that is, the QL value, and the value of the known spectral reflectance from the image of the Macbeth chart whose spectral reflectance is simultaneously known when the multiband image is captured. From the created one-dimensional look-up table (hereinafter referred to as a LUT) 20 and the QL value which is the image data value of the original image obtained by the data processing unit 18, the spectral reflectance is calculated using the one-dimensional LUT 20. A spectral reflectance spectrum calculating unit 22 that converts the spectral reflectance into a value and estimates a spectral reflectance, and a noise removal process that removes a noise component of the obtained spectral reflectance value 24, wavelength shift processing unit 25 for adjusting a shift amount of the wavelength correcting filter temperature to be described later
An interpolation processing unit 26 that linearly interpolates between the wavelengths of the spectral reflectance obtained by the spectral reflectance spectrum calculation unit 22 to increase the number of channels;
4, a compression / expansion processing unit 2 that compresses or expands the wavelength of the spectral reflectance obtained by the spectral reflectance spectrum calculation unit 22 in accordance with the half-width that is a characteristic of the bandpass filter.
8, among the Macbeth charts taken at the same time when the multi-band image is taken, the saturation is high and the spectral reflectance is sharp with respect to the wavelength, such as red (R), green (G) and blue (B). And a filter for correcting the temperature of the liquid crystal tunable filter 14, which includes a spectral reflectance of the liquid crystal tunable filter 14. The sharpness degradation restoration processing (deconvolution processing) of the spectral distribution of the spectral reflectance that has become dull due to the overlap between the channels having a half-value width, which is the characteristic of the liquid crystal tunable filter 14 as a bandpass filter, is performed. , A deconvolution processing unit 34 that obtains a spectrum of the calibrated spectral reflectance, and a wave that has been In response to compression or decompression of mainly and a decompression and compression inverse transform unit 36 to restore the wavelength of the spectrum of the calibrated spectral reflectance stretched or compressed into.

The components constituting the spectral reflectance spectrum estimating apparatus 16 in this embodiment are executed by a computer as software functions. However, the present invention is not limited to software, and some or all of the components may be implemented. The functions may be implemented in hardware.

The data processing section 18 performs A / D conversion of an output signal obtained by receiving light by the CCD camera 12, obtains digital image data, corrects DC offset, and then applies darkness correction. Thereafter, the image data is subjected to LOG conversion, shading correction is performed, and a QL value of the image data of the original image forming the multi-band image is obtained.
The QL value of the obtained original image is sent to the spectral reflectance spectrum calculation unit 22.

The one-dimensional LUT 20 stores a C
Using a CD camera 12 and a liquid crystal tunable filter 14, a Macbeth chart is photographed as a multi-band image having the same number of channels and a wavelength region at the time of photographing the subject, and the spectral reflectance of the sixth stage of the obtained Macbeth chart image is obtained. Is a portion provided with a Macbeth one-dimensional LUT that associates the QL value of a known gray patch with the known spectral reflectance. When the spectral reflectance spectrum calculating section 22 calculates the spectral reflectance, The LUT created by the LUT 20 is referred to. The chart used in the present invention is not limited to the Macbeth chart, and may be any chart as long as its spectral reflectance is known.

The spectral reflectance spectrum calculator 22 converts the QL value of the image data of the original image of the subject obtained by the data processor 18 into a spectral reflectance value with reference to the one-dimensional LUT 20. This is the part for obtaining the spectral reflectance. 1
If the reference data does not exist in the dimensional LUT 20, the value of the spectral reflectance is obtained by interpolation or extrapolation, but the method of interpolation or extrapolation is not particularly limited. For example, known methods such as linear interpolation and Lagrange interpolation can be used. The obtained data of the spectral reflectance is sent to the noise removal processing unit 24.

The noise removal processing section 24 is a processing section for removing a noise component contained in the original image. In the original image of the short wavelength side channel, the noise removal strength is increased, and the original image of the long wavelength side channel is removed. The image is configured to weaken the strength of noise removal. With such a configuration, the spectral transmittance of the liquid crystal tunable filter 14 is low in the channel on the short wavelength side as shown in FIG. 2, a sufficient amount of light cannot be secured, and noise is easily picked up. On the other hand, in the channel on the long wavelength side, the transmittance is high, a sufficient amount of light can be secured, and it is difficult to pick up noise components. The method for removing noise is not particularly limited. For example, there is a method of automatically using a median filter on-off when the peak value of the spectral transmittance becomes 0.1 or less. The spectral reflectance data from which noise has been removed is sent to the wavelength shift processing unit 25.

The wavelength shift processing unit 25 adjusts the temperature correction of the spectral reflectance data from which noise has been removed due to the temperature of the liquid crystal tunable filter 14 by the wavelength shift amount. The wavelength shift processing is performed using the shift amount Δλ obtained by the unit 32. After that, it is sent to the interpolation processing section 26.

The interpolation processing unit 26 calculates the spectral reflectance of the spectral reflectance obtained by the spectral reflectance spectrum calculating unit 22 in order to obtain the spectral distribution of the spectral reflectance finally estimated as a smooth curve without unevenness. Is interpolated non-linearly by spline interpolation, Lagrange interpolation, or the like, and the data of the spectral distribution of the spectral reflectance is reduced to, for example, a wavelength interval of 10 nm or less. Further, the number of channels may be increased by linearly interpolating the data of the spectral distribution of the spectral reflectance. In this case, the spectral distribution of the spectral reflectance calibrated by the Fourier transform and the inverse Fourier transform, which will be described later, is a smooth curve even if the curve is uneven by linear interpolation. The spectral distribution of the spectral reflectance obtained by making the data fine by the interpolation is sent to the compression / expansion conversion unit 28.

The compression / expansion processing converter 28 changes the wavelength of the spectral distribution of the spectral reflectance obtained by the spectral reflectance spectrum calculator 22 with the change of the wavelength of the liquid crystal tunable filter 14 as shown in FIG. This is a portion that compresses or expands the wavelength of the spectrum of the spectral reflectance of the subject according to the half width of the spectral distribution of the spectral transmittance to be performed. In other words, the spectral reflectance of the spectral reflectance obtained by the spectral reflectance spectrum calculator 22 is determined by the ratio of the half-width of the spectral transmittance and the half-width of the spectral distribution, which varies with the wavelength of the liquid crystal tunable filter 14, to the half-width of the reference. Compress or expand the wavelength. The reference half-value width is, for example, the center wavelength of a wavelength region covered by all channels at the time of shooting, for example, in the example of FIG. 2, the center wavelength of 550 nm of a wavelength region of 410 to 710 covered by 16 channels. The half-width of the spectral transmittance distribution spectrum included in the range of 50 nm from the center wavelength, for example, the spectral transmittance distribution spectrum A in the example of FIG. As described above, changing the wavelength of the spectral distribution of the spectral reflectance according to the half-value width of the spectral transmittance is performed by using a device function that changes according to the wavelength, which is used in a deconvolution process described later, for each channel according to the wavelength. This is because the deconvolution process can be easily performed using one device function.
The magnification of compression or expansion varies depending on the wavelength. For example, in the case of having a spectral transmittance distribution as shown in FIG. 2, the compression ratio is high because the half width is large on the high wavelength side, and the half width is small on the low frequency side. Conversely, the expansion ratio is high, and in this region, the compression ratio and the expansion ratio change smoothly in accordance with the wavelength.

On the other hand, the temperature correction spectral reflectance data section 30
Is a part for calculating the spectral distribution of the spectral reflectance when a patch having a known spectral reflectance is photographed using the liquid crystal tunable filter 14. As described later, the change in the liquid crystal temperature of the liquid crystal tunable filter 14 It is used to determine the amount of wavelength shift of the spectral transmittance distribution that changes according to. The temperature correction spectral reflectance data section 30 is a one-dimensional L
Using the image of the Macbeth chart used when creating the Macbeth gray one-dimensional LUT with the UT 20, the spectral distribution of the spectral reflectance of the image of the patch of the R, G, and B hues of which the spectral reflectance is known among the images of the Macbeth chart Using a Macbeth Gray one-dimensional LUT. The spectral distribution of the spectral reflectance of the R, G, and B hue patches is determined because the saturation is high and the spectral reflectance spectral distribution changes rapidly with wavelength. Because you can ask. In addition, Macbeth Gray 1
The image of the Macbeth chart used when creating the dimensional LUT or when using the spectral reflectance data unit for temperature correction 26 may be taken together with the subject at the time of shooting the subject. Further, the chart image used in the present invention is not limited to the Macbeth chart image, and has a patch whose spectral reflectance is known, the chroma is high, and the spectral distribution of the spectral reflectance changes abruptly with respect to the wavelength. Any image of the chart may be used.

The filter temperature correction unit 32 calculates the spectral distribution of the spectral reflectance of the R, G, and B hue patches obtained by the temperature correcting spectral reflectance data unit 30 and the known spectral reflectance of the patch. This is a part for obtaining a shift amount of the waveform due to the temperature by comparing with the spectrum distribution. The shift amount obtained by the filter temperature correction unit 32 is used when selecting a wavelength shift amount Δλ used in the wavelength shift processing performed by the wavelength shift processing unit 25, as described later. The obtained shift amount of the waveform due to the temperature is sent to the wavelength shift processing unit 25.

The spectral distribution of the spectral reflectance processed by the interpolation processing unit 26 is changed by the compression / expansion conversion unit 28 in accordance with the wavelength characteristic of the spectral transmittance of the liquid crystal tunable filter 14 in accordance with an operator's instruction. The deconvolution processing unit 3 performs no compression / expansion processing to be performed.
4 may be sent. Although not shown, the spectral distribution of the spectral reflectance may be sent directly from the spectral reflectance spectrum calculator 22 to the deconvolution processor 34 and from the noise removal processor 24 to the deconvolution processor 34.

The deconvolution processing section 34 calibrates the spectral distribution of the spectral reflectance of the subject having a dull waveform estimated by the spectral reflectance spectrum calculating section 22 using an apparatus function, and obtains the calibration based on the characteristics of the measuring apparatus. This is the part for obtaining the spectral distribution O (λ) of the spectral reflectance unique to the subject. That is, as shown in FIG. 2, the spectral transmittance distribution of the liquid crystal tunable filter 14 overlaps with the spectral transmittance distribution located in the vicinity of the spectral transmittance distribution due to the spread of the spectral transmittance distribution. The spectral distribution of the spectral reflectance of the subject estimated by the spectral reflectance spectrum calculating unit 22 has a dull waveform. Therefore, in order to obtain a clearer spectral distribution, the spectral transmittance distribution of the liquid crystal tunable filter 14 and the spectral sensitivity distribution of the camera that captures the image (the spectral sensitivity distribution including the spectral transmittance of the optical system).
And the apparatus function h (λ) determined by the product of the spectral intensity distribution of the photographing illumination at the time of photographing.

The device function h (λ) is represented by the spectral transmittance distribution of the liquid crystal tunable filter 14 and the spectral sensitivity distribution of the camera for photographing (the spectral sensitivity distribution including the spectral transmittance of the optical system).
And the spectral intensity distribution of the photographing illumination at the time of photographing. The spectral transmittance distribution of the liquid crystal tunable filter 14 depends on the wavelength, that is, the channel, as shown in FIG.
In the example of the figure, the spectral transmittance distribution A having a value within the range of 50 nm from the center wavelength 550 nm of the wavelength region 410 nm to 710 nm covered by all the channels is taken as a representative function, A device function h (λ) defined by the product of the spectral sensitivity distribution of the camera (the spectral sensitivity distribution including the spectral transmittance of the optical system) and the spectral intensity distribution of the imaging illumination at the time of imaging is set. And
This apparatus function h (λ) function and the spectral distribution I (λ) of the spectral reflectance of the subject having a dull waveform estimated by the spectral reflectance spectrum calculating unit 22 are Fourier-transformed to obtain F [h (λ )] And F [I (λ)] (F
[] Represents the Fourier transform), and from this,
A division such as the following equation (4) is performed, and then an inverse Fourier transform is performed to obtain the spectral distribution O (λ) of the spectral reflectance specific to the subject irrespective of the characteristics of the liquid crystal tunable filter 14 and the CCD camera 12. obtain. O (λ) = F ⁻¹ [(F [I (λ)] / F [h (λ)]]] (4) The Fourier transform method is an FFT (Fast Fourier Transfo
rmation) or a method by a known method such as a method by Fourier expansion.

As described above, the deconvolution is calculated by the equation (4).
As described later, the spectral processing of the subject is performed.
The spectral distribution I (λ) of the emissivity is
Multi-band image by filter 14, CCD camera 12, etc.
The characteristics of the device that captures the image, for example,
Depending on the spectral transmittance distribution and the like of the
Because the spectral distribution O (λ) of the spectral reflectance becomes dull.
The spectral distribution I of the spectral reflectance of the dull subject
(Λ) is a device function which is a characteristic of the device as described later.
h (λ) to calibrate by removing device characteristics
Spectra of the spectral reflectance specific to the subject that has restored the opacity degradation
This is for obtaining the torque distribution O (λ). This Fourier transform
The deconvolution processing method according to
The processing is performed using one device function h (λ) instead of being provided for each channel.
In order to perform the processing, the conventional matrix-based method using equation (2)
The known spectral transmittance distribution f of the tunable filter
_i(Λ) (i = 1 to n) is stored for each channel, and
There is no need to arrange device functions for each channel, and device functions h
Since only one (λ) is prepared, the processing becomes simple and the total
Calculation time is reduced.

The method of deconvolution processing performed by the method for estimating the spectral reflectance of a multi-band image according to the present invention is a method of obtaining a spectral distribution O (λ) of a spectral reflectance specific to an object by equation (4). In addition to the above, a function F [h obtained by Fourier-transforming the device function h (λ)
(Λ)] on the Fourier plane is calculated as a reciprocal, and a convolution integral of a waveform obtained by performing an inverse Fourier transform thereof and the spectral distribution I (λ) of the spectral reflectance of the subject is calculated (convolution processing), and the uniqueness of the subject is calculated. The spectral distribution O (λ) of the spectral reflectance may be obtained.

In the above embodiment, the deconvolution processing method by the Fourier transform is used. However, the present invention is not limited to this, and the conventional calculation using the matrix by the equations (2) and (3) can be performed. May be used,
From the known spectral transmittance distribution f _i (λ) (i = 1 to n) for each channel of the wavelength tunable filter, spectral transmittance having a value within a range of 50 nm above and below the center wavelength of the wavelength region covered by all channels. The power distribution is defined as a function of a representative spectral transmittance distribution, and this function and the spectral sensitivity distribution of the camera that captures the image (the product of the spectral sensitivity distribution S (λ) of the CCD camera and the known spectral transmittance L (λ) of the optical system) And the product of the spectral intensity distribution E (λ) of the photographing illumination at the time of photographing, and the deconvolution processing may be performed by the equations (2) and (3). As a result, the matrix calculation of Expressions (2) and (3) can be easily performed, and the calculation time can be reduced.

In particular, the number of channels is increased in the interpolation processing unit 26 so that the number n of channels is equal to the number m of measurement points of the spectral distribution I (λ) of the spectral reflectance of the subject, and the spectral reflectance of the subject is Spectral distribution I (λ)
May be made equal to the number of measurement points m, and an inverse matrix of the matrix F of the equation (2) may be obtained. This makes it possible to analytically and in a short time obtain the inverse matrix of the matrix F in the equation (2). The obtained spectral distribution of the spectral reflectance specific to the subject is sent to the expansion / compression inverse converter 36.

The expansion / compression inverse converter 36 performs inverse conversion of wavelength expansion / compression by multiplying the inverse magnification so as to cancel the wavelength compression / expansion processing performed by the compression / expansion converter 28. Part. That is, in the compression / expansion conversion unit 28, the wavelength of the spectral distribution of the spectral reflectance is changed according to the half width of the spectral transmittance, and then the deconvolution process is performed, so that the calibrated spectral reflectance of the deconvolution process is The wavelengths of the spectral distribution still remain compressed or expanded. for that reason,
In order to return the spectral distribution of the spectral reflectance specific to the subject to the original wavelength, the expansion or compression is performed by multiplying the compression or expansion performed by the compression / expansion converter 28 by a reverse magnification so as to cancel the compression or expansion. Do. In this way, the inverse transform of expansion or compression can be performed to obtain a calibrated spectral reflectance spectral distribution of the subject. The obtained spectral distribution of the calibrated spectral reflectance of the subject is stored in a memory storage unit (not shown).

The spectral reflectance spectrum estimating device 16 is configured as described above. Next, a method for estimating the spectral reflectance spectrum of a multiband image according to the present invention will be described.

FIG. 3 shows an outline of an example of a method for estimating the spectral reflectance of a multiband image. As shown in FIG. 3, a method of estimating the spectrum of the spectral reflectance of a multiband image is performed by, for example, photographing a subject with a multiband camera including a liquid crystal tunable filter 14 and a CCD camera 12, and setting the specific wavelength region. Image data of a multi-band image composed of original images for each channel, that is, digital data values in pixel units that have been AD-converted in the data processing unit 18, that is, QL values, for example, integer values of 0 to 255 in 8-bit AD conversion are obtained.

On the other hand, at the time of photographing the subject or before and after the photographing of the subject, a known Macbeth chart of four rows and six rows was photographed by the multi-band camera, and the spectral reflectance data of the Macbeth chart was obtained as a QL value. Then, the QL value of the gray tone without hue in the fourth row of the Macbeth chart is
A Macbeth gray one-dimensional LUT is created in correspondence with the spectral reflectance of a known gray gradation known in advance. Using the created Macbeth Gray one-dimensional LUT, the spectral distribution of the spectral reflectance of the subject is estimated from the QL value of the multi-band image of the subject.

The spectral distribution of the estimated spectral reflectance of the subject is subjected to noise removal processing, interpolation interpolation processing and compression / expansion conversion as necessary, and furthermore, from the spectral reflectance data of the Macbeth chart, the If necessary, filter temperature correction (wavelength shift) is performed using the data of the spectral reflectance of the R, G, and B patches whose spectral reflectance changes rapidly with wavelength, and then the spectral The sharpness deterioration restoration processing (deconvolution processing) and the inverse expansion / compression conversion are performed as necessary to obtain the spectral distribution of the calibrated spectral reflectance.

That is, in the method for estimating the spectral reflectance of a multi-band image according to the present invention, the spectral distribution of the spectral reflectance of the subject is determined from the QL value of the multi-band image of the subject using a Macbeth Gray one-dimensional LUT. This method is based on estimation, and is a method of performing deconvolution processing, interpolation, compression / decompression processing, and temperature correction as needed to obtain a more accurate spectral distribution.

FIG. 4 is a flowchart showing an example of a flow described for each process based on the multiband image spectral reflectance spectrum estimation system 10 for implementing the spectrum estimation method shown in FIG.

First, a subject is photographed by a multi-band camera including a liquid crystal tunable filter 14 and a CCD camera 12 in which a long region is divided into 16 channels for each specific wavelength region as shown in FIG. The image signal of a multi-band image composed of original images for each specific wavelength region set by the filter 14 is obtained, and the data processing unit 18 of the spectral reflectance spectrum estimating device 16 performs AD conversion of 8 bits to obtain a DC offset. Correction, darkness correction, LOG conversion, and shading correction are performed to obtain a QL value consisting of an integer of 0 to 255 for each pixel unit of the original image corresponding to 16 channels, and to obtain image data of a multi-band image (step). 10
0).

On the other hand, a Macbeth chart having a known spectral reflectance is photographed simultaneously with or before or after the multiband image is photographed in step 100, and the spectral luminance of the Macbeth chart is measured (step 102). Here, the spectral luminance measurement means that the data of the multi-band image signal of the Macbeth chart is subjected to 8-bit AD conversion in the data processing unit 18 of the spectral reflectance spectrum estimating apparatus 16 in the same manner as the acquisition of the multi-band image data in step 100. , DC offset correction, darkness correction, LOG conversion, and shading correction to obtain a QL value corresponding to each patch of the Macbeth chart. Of the QL values corresponding to each patch of the Macbeth chart thus obtained,
A one-dimensional LUT is created in which the QL values of the six gray patches in the fourth row at the bottom and the known spectral reflectance values of the gray patches correspond one-to-one to each original image (step 104). ). The one-dimensional L created here
The UT includes correction of the transmittance level of each wavelength of the liquid crystal tunable filter 14.

Next, the QL value of the original image corresponding to each channel of the photographic subject obtained in step 100 is converted into a spectral reflectance value using the one-dimensional LUT created in step 104 (step 106). . QL value to refer to is 1
If it does not exist in the dimensional LUT, the value of the spectral reflectance is obtained by interpolation or extrapolation. The method of interpolation and extrapolation is not particularly limited, and is performed by a known method, for example, a known method such as linear interpolation or Lagrange interpolation.

As described above, the spectral reflectance can be converted to the value of the spectral reflectance using the one-dimensional LUT, as shown in FIG. The distribution is different from the conventional filter in that the wavelength resolution is improved, a steep distribution centering on the peak value is formed, the wavelength overlap between the channels is reduced, and the spectral reflectance obtained using the one-dimensional LUT is reduced. This is because the bluntness of the spectrum distribution is reduced.
Explaining the equation (2), since the off-diagonal components of the matrix F shown in the equation (2) are almost zero, the matrix F can be converted to a value of spectral reflectance using a one-dimensional LUT. Therefore, it is not necessary to calculate the estimation matrix G as in the equation (3), and the processing time is greatly reduced.

On the other hand, in order to perform a filter temperature correction described later, patches of R, G, and B hues obtained in the Macbeth chart spectral luminance measurement in step 102 are used by using the one-dimensional LUT obtained in step 104. The spectral distribution of the spectral reflectance is calculated (step 108).

Next, the filter temperature is corrected (step 109). The filter temperature correction is performed by using a device function h (λ) selected as a representative function when a deconvolution process (step 118) described later is performed, based on the spectral transmittance distribution of the liquid crystal tunable filter 14 and the camera to be photographed. The spectral transmittance distribution of the liquid crystal tunable filter 14 is defined by the product of the spectral sensitivity distribution (spectral sensitivity distribution including the spectral transmittance of the optical system) and the spectral intensity distribution of the shooting illumination at the time of shooting. This is because the spectral transmittance distribution shifts in the wavelength region depending on the temperature of the portion. In the filter temperature correction, in step 108, the spectral reflectance distribution obtained by measuring patches of R, G, and B hues whose spectral reflectance distributions are known, and the known spectral reflectance distribution are calculated. By comparison, the shift amount of the wavelength is obtained. R, G
And measuring the hue patches of B are highly saturated,
This is because the spectral reflectance distribution changes abruptly with respect to the wavelength, so that the shift amount of the wavelength can be easily obtained.

In the above embodiment, the amount of wavelength shift is determined from the difference between the spectral reflectance distribution of a patch having a known spectral reflectance distribution and the spectral reflectance distribution obtained by photographing the patch with a multi-band camera. This is used to correct the spectral reflectance, but the present invention is not limited to this method, and a thermometer is installed so as to be in contact with the liquid crystal portion of the liquid crystal tunable filter 14 so that the temperature of the liquid crystal portion can be measured during multi-band image capturing. Alternatively, the data may be recorded through a filter temperature acquiring unit of the spectral reflectance spectrum estimating device 16 (not shown), and the wavelength shift amount corresponding to the temperature may be obtained using the recorded temperature to correct the spectral reflectance.

On the other hand, with respect to the spectral reflectance data of the original image corresponding to each channel obtained in step 106,
A noise removal process is performed (step 110). In the noise removal processing, as shown in FIG. 2, the spectral transmittance of the liquid crystal tunable filter 14 is low in the channel on the short wavelength side, a sufficient amount of light cannot be secured, and noise is easily picked up. Since the channel on the wavelength side has a high transmittance, a sufficient amount of light can be secured, and it is difficult to pick up noise components, a noise removal process that can adjust the strength of noise removal according to the wavelength region of each channel is desirable. For example, there is a method of automatically using the median filter on-off when the peak value of the spectral transmittance becomes 0.1 or less.

Next, after the noise removal processing is performed on the data of the spectral reflectance in step 110,
The wavelength shift process is performed by correcting the wavelength of the spectral reflectance that changes due to this wavelength shift by Δλ using the wavelength shift amount Δλ obtained in step 9 (step 11).
1).

Further, the spectral reflectance field data subjected to the noise removal processing obtained in step 110 is subjected to interpolation processing (step 112). The reason why the interpolation is performed is to finally increase the number of channels in the spectral reflectance spectral distribution to obtain a smooth distribution. As the interpolation, an interpolation process is performed by a non-linear interpolation such as a spline interpolation or a Lagrange interpolation so as to have a spectrum distribution of, for example, 10 nm or less. Also, by linear interpolation,
You may obtain a spectral distribution of spectral reflectance with unevenness,
In this case, a deconvolution process described later performs a process using a Fourier transform and its inverse transform. This is because a smooth spectral distribution can be obtained even with an uneven spectral distribution by the Fourier transform and its inverse transform.

A compression / expansion process for compressing or expanding the wavelength of the spectral distribution of the spectral reflectance of which data has been made fine by performing the interpolation process (step 112) is performed (step 114). The compression / expansion processing changes the wavelength of the spectral distribution of the spectral reflectance according to the half width of the spectral distribution of the spectral transmittance of the liquid crystal tunable filter 14 which changes with the change of the wavelength shown in FIG. Compress or expand the wavelength of the spectrum. The ratio of compression and expansion changes for each wavelength range so that the wavelength range of the spectral reflectance spectrum of the subject becomes constant regardless of the channel with respect to the half-value width of one device function h (λ) described later. I do.
As a result, in the deconvolution processing described later, it is not necessary to have a plurality of device functions h (λ) defined by the spectral transmittance distribution that changes according to the wavelength, and the deconvolution process is performed using only one device function. be able to.

The spectral distribution of the spectral reflectance subjected to the compression / expansion conversion (step 114) in this way is subjected to a deconvolution process for restoring the sharpness degradation of the spectral reflectance spectral distribution (step 11).
8). Originally, liquid crystal tunable filter 14 and CC
The spectral distribution of the spectral reflectance of the subject obtained from the multi-band camera provided with the D camera 12 is in a filtered state according to the characteristics of the device of the multi-band camera that has taken the image. Therefore, the spectral distribution is not sharp but obtained in a dull form. On the other hand, since the device function h (λ) which is a characteristic of the device of the photographed multi-band camera is known, the sharpness deterioration of the spectral distribution of the dull spectral reflectance is restored using the device function h (λ). To process. That is, as shown in FIG.
The spectral distribution O (λ) of the spectral reflectance specific to the subject is
The device function h (λ) is decomposed for each channel, multiplied by a device function h _i (λ) determined for each channel, and then the distribution calculated by summing the entire channel is represented by a multi-function including the device function h (λ). The spectral distribution I (λ) of the spectral reflectance captured by the band camera is obtained. As described above, the spectral distribution I (λ) of the measured spectral reflectance is obtained.
Is the spectral distribution O (λ) of the spectral reflectance specific to the subject.
And the device function h (λ) are determined by a convolution process (convolution process) of the following equation (5).

(Equation 4)

The deconvolution processing performed in step 118 uses the spectral distribution I (λ) of the spectral reflectance obtained by the above convolution and the device function h (λ), and conversely to the convolution processing, This is a process for obtaining the spectral distribution O (λ) of the spectral reflectance. The convolution integral shown in equation (5) is
As shown in the following equation (6), F [I
(Λ)], F [O (λ)] and F [h (λ)]. Therefore, as in equation (4), F [I
(Λ)] is divided by F [h (λ)] and inverse Fourier-transformed to obtain O (λ).

(Equation 5)

When the number of pixels of the multiband image is small,
For example, when the number of pixels is less than 100,000 pixels, in the deconvolution processing by the Fourier transform, all the device functions corresponding to each channel, for example, 16 channels are stored, and it is not necessary to perform an operation corresponding to each channel by using this. Since the calculation can be performed using only the device function h (λ) of one channel, the processing of the spectral distribution of the spectral reflectance specific to the subject can be performed in a short time.

When the number of pixels of the multi-band image is large, for example, when the number of pixels is 100,000 or more, the deconvolution processing can be performed by using a conventional matrix-based calculation using equations (2) and (3). At this time, the known spectral transmittance distribution f for each channel of the tunable filter is often used.
_{From i} (λ) (i = 1 to n), the spectral transmittance distribution of a specific channel having a value within a range of 50 nm above and below the center wavelength of the wavelength region covered by all channels is represented by a representative spectral transmittance distribution. Function and the spectral sensitivity distribution of the camera that captures the image (the spectral sensitivity distribution S of the CCD camera).
(Product of (λ) and known spectral transmittance distribution L (λ) of optical system)
And the product of the spectral intensity distribution E (λ) of the photographing illumination at the time of photographing, and the deconvolution processing by the equations (2) and (3) may be performed. As a result, the matrix calculation of Expressions (2) and (3) can be easily performed, and the calculation time can be reduced.

In particular, the number of channels is increased in the interpolation process (step 112) so that the number of channels n is equal to the number of measurement points m of the spectral distribution I (λ) of the spectral reflectance of the subject, and the spectral reflectance of the subject is increased. The number of measurement points m of the rate spectral distribution I (λ) may be made equal to obtain the inverse matrix of the matrix F in the equation (2). As a result, the inverse matrix of the matrix F in the equation (2) can be obtained analytically and in a short time.

Deconvolution processing (step 11)
After 8) is performed, the expansion / compression inverse conversion, which is the inverse of the compression / expansion processing performed in step 114, is performed (step 120). As a result, the calibrated spectral reflectance spectral distribution compressed or expanded in the wavelength direction in step 114 is obtained (step 122) and stored in the storage unit. In this way, it is possible to obtain a calibrated spectral reflectance spectral distribution from which the characteristics of the multiband camera have been removed.

Next, based on the method and system for estimating the spectral reflectance of the spectral reflectance of the multiband image, the spectrum of the spectral reflectance of the subject was obtained from the multiband image. DA as a CCD camera
LSA CA-D4-1024A (1024 x 102 pixels)
4. A Varispec Tunable Filter (liquid crystal tunable filter) manufactured by CRI was used as a wavelength variable filter using a pixel size of 12 × 12 microns, with a PCI interface, and monochrome. In this liquid crystal tunable filter, the center wavelength of the spectral transmittance distribution can be arbitrarily selected within the filter wavelength range of 400 to 720 nm, and the average value of all the channels having a half width of the distribution wavelength is 30 nm.
And the spectral transmittance varied from 6 to 60% depending on the wavelength. PRO as spectral reflectance spectrum estimator
Book type PC (personal computer) manufactured by SIDE
, Software for estimating the spectral distribution of the spectral reflectance on Windows 95 was created in C ⁺⁺ language.
The book type PC manufactured by PROSIDE has a CPU of 1
It was 66 MHz and the RAM was 128 Mbytes.

A metal halide lamp was used as a light source for photographing, and the illuminance of the photographing subject was set to 12,000 lux. The shooting lens used for shooting is Nikomart (f = 50mm, F
1.4), and an ultraviolet cut filter of 400 nm or less and an infrared cut filter of 730 nm or more were used. The imaging wavelength range of 410 nm to 710 nm is set to 16
It was divided into 16 channels, and each wavelength interval was 20 nm. For photographing, a human face and a Macbeth chart were simultaneously photographed. In addition, shooting by changing the spectral transmittance distribution of the liquid crystal tunable filter according to each divided channel,
Sixteen original images constituting the multiband image were obtained. The aperture value at the time of shooting was F2.8. The shooting time was 25 ms for one shot, and it took 3 seconds for the entire shooting of the multiband image.

First, as a process a, image data values (QL values) of six gray patches whose spectral reflectances are known at the bottom of the Macbeth chart image in the original image corresponding to each channel are obtained. Is obtained for each channel, a QL value of this gray patch image is associated with a known spectral reflectance value of this gray patch, and a one-dimensional lookup table is created for each channel. From the QL value, the spectral distribution of the spectral reflectance of the human face of the photographing subject was obtained using the one-dimensional lookup table created earlier.

As processing b, before photographing the multi-band image, the spectral sensitivity distribution of the CCD camera (the spectral sensitivity distribution including the spectral transmittance of the optical system) and the liquid crystal tunable in the wavelength range corresponding to each channel. The spectral transmittance distribution of the filter was measured, and then the process a was performed. Furthermore, the spectral transmittance distribution of the wavelength range corresponding to the channel of the liquid crystal tunable filter, the spectral sensitivity distribution of the CCD camera, and the known spectral intensity distribution of the illumination light at the time of shooting are multiplied by a specific channel. Using the device function h (λ) and the spectral distribution I (λ) of the spectral reflectance of the human face of the photographing subject, deconvolution by the Fourier transform method using FFT according to the equation (4), using the device function h (λ) The process was performed to obtain a calibrated spectral distribution of the spectral reflectance specific to the subject.

As the process c, as in the case of the process b, before photographing the multi-band image, the spectral sensitivity distribution of the CCD camera (the spectral sensitivity distribution including the spectral transmittance of the optical system) and each channel are corresponded. The spectral transmittance distribution of the liquid crystal tunable filter in the wavelength range was measured, and then the process a was performed. Then, the spectral sensitivity distribution of the CCD camera (the spectral sensitivity distribution including the spectral transmittance of the optical system), the spectral transmittance distribution of the liquid crystal tunable filter in the wavelength range corresponding to each channel, and the known illumination light at the time of shooting. The matrix F of the equation (2) was calculated from the spectral intensity. At that time, the number of measurement points of the multiband image was 16 (n = 16), and the wavelength interval was 20 nm. On the other hand, the wavelength interval between the spectral sensitivity distribution of the CCD camera and the spectral transmittance distribution in the wavelength region of the liquid crystal tunable filter is 5 nm, and the finally obtained spectral distribution of the spectral reflectance specific to the calibrated subject is also 5n.
m, and the number of channels was 81 (m = 81) in the imaging wavelength range of 410 nm to 710 nm. Therefore, after multiplied by the transposed matrix F ^t of the matrix F in the both sides of the equation (2), the matrix (F ^t · F) determined the inverse of the value of the right side object-specific spectral reflectance of the formula (2) O _k (k = 1 to 81) was calculated, and the spectral distribution O (λ) of the calibrated spectral reflectance was calculated and obtained. The calculation for obtaining the inverse matrix used the Gauss-Seidel method.

The spectral distribution of the obtained spectral reflectance is converted into color space coordinates, and the average value (ΔR _av ) and the average value (ΔE _av ) of the absolute values of the differences between the spectral reflectances of the 24 patches of the Macbeth chart are calculated. I asked. Here, the difference in the spectral reflectance is determined because not only the color difference but also the accuracy of the spectral reflectance becomes a problem. Even when the color difference is 0, a large difference in spectral reflectance may occur, that is, a phenomenon such as metamerism may occur. In the processing a, the processing time required for the processing a is 1 minute and 15 seconds, ΔR _av is 0.0093, ΔE
_av was 2.70. In the processing b, the processing time required for the processing b is 13 minutes 01 seconds, ΔR _av is 0.0083,
ΔE _av was 1.68. Further, the processing time required for the processing c is 5 minutes and 36 seconds, ΔR _av is 0.0080, Δ
E _av was 1.67.

As described above, the processes a, b and c can be performed in a short time without effectively reducing the average value ΔR _av and the average value ΔE _av of the absolute value of the difference between the spectral reflectances. it is obvious.

Although the method and system for estimating the spectral reflectance of a multiband image according to the present invention have been described in detail above, the present invention is not limited to the above-described embodiment, and does not depart from the gist of the present invention. At
Of course, various improvements and changes may be made.

[0070]

As described in detail above, according to the present invention, when a multi-band image is photographed using a wavelength tunable filter and the spectrum of the spectral reflectance of the photographed subject is estimated using the photographed image. In addition, the processing time for obtaining the estimated spectrum can be reduced without effectively lowering the spectrum estimation accuracy.

[Brief description of the drawings]

FIG. 1 is a block diagram schematically illustrating an example of a system for estimating a spectral reflectance of a multiband image according to the present invention.

FIG. 2 is a diagram showing an example of the spectral transmittance of the liquid crystal tunable filter shown in FIG.

FIG. 3 is an explanatory diagram illustrating an outline of an example of a method of estimating a spectral reflectance spectrum of a multiband image according to the present invention.

FIG. 4 is a flowchart illustrating a flow of an example of a method for estimating a spectral reflectance spectrum of a multiband image according to the present invention.

FIG. 5 is an explanatory diagram illustrating a deconvolution process shown in FIG. 4;

[Explanation of symbols]

Reference Signs List 10 Spectrum estimation system for multiband image spectral reflectance 12 CCD camera 14 Liquid crystal tunable filter 16 Spectral reflectance spectrum estimating device 18 Data processing unit 20 One-dimensional LUT 22 Spectral reflectance spectrum calculating unit 24 Noise removal processing unit 26 Interpolation Processing unit 28 Compression / expansion conversion unit 30 Spectral reflectance data unit for temperature correction 32 Filter temperature correction unit 34 Deconvolution processing unit 36 Expansion / compression inverse conversion unit

Claims (13)

[Claims] An imaging wavelength region is divided into a plurality of channels by using a wavelength variable filter, and a multi-band image composed of a plurality of original images obtained by imaging the same subject by an imaging means corresponding to each channel is obtained. A method for estimating a spectrum of a spectral reflectance of the subject from a multi-band image, wherein the wavelength variable filter has 10 or more channels, and an average wavelength half width of a spectral transmittance distribution of all channels is 40 nm. A wavelength tunable filter, which is as follows: a conversion table in which a QL value obtained by photographing a chart having a known spectral reflectance is associated with the spectral reflectance for each channel, and a conversion table is created for each of the multiband images. By converting the QL value of the original image of the channel into the spectral reflectance using the conversion table, the spectral reflectance of the subject is obtained. Spectral estimation method of the spectral reflectance of the multi-band images and estimates the spectrum. 2. The method according to claim 1, wherein the variable wavelength filter is a liquid crystal tunable filter. 3. A spectral transmittance distribution of the tunable filter in a wavelength region of the channel, a spectral sensitivity distribution of the photographing unit, and a spectral sensitivity distribution of the estimated spectral reflectance of the subject. 3. The spectrum of a multiband image according to claim 1 or 2, wherein a deconvolution process is performed to calibrate the spectrum of the spectral reflectance of the subject using one device function determined by a product of the spectral intensity distribution of the photographing illumination light. Method for estimating reflectance spectrum. 4. The method according to claim 3, wherein the deconvolution processing is performed by using a Fourier transform or a matrix operation. 5. An apparatus function having a value within a range of 50 nm above and below a center wavelength of a wavelength region photographed by the channel, as the apparatus function.
6. The method for estimating the spectral reflectance of a multiband image according to the above. 6. The deconvolution process according to claim 1, wherein the wavelength of the spectral reflectance spectrum of the object is compressed or expanded according to the half-value width of the tunable filter before performing the deconvolution process. later,
The multiband image according to claim 3, wherein inverse conversion is performed to expand or compress the wavelength of the spectrum of the spectral reflectance of the subject obtained by the deconvolution processing in accordance with the compression or expansion of the wavelength. Method for estimating the spectral reflectance of the spectrum. 7. The spectrum estimation of a spectral reflectance of a multiband image according to claim 2, wherein a filter temperature correction for compensating for a temperature dependency of the wavelength tunable filter is performed when the deconvolution processing is performed. Method. 8. The filter temperature correction according to claim 1, wherein the spectral reflectance distribution is known, and a high-saturation color chart showing a sharp change with respect to wavelength is photographed. A shift amount of a wavelength at which the obtained waveform of the spectral reflectance distribution coincides with the known waveform of the spectral reflectance is determined, and correction of the spectral reflectance distribution of the subject is performed according to the obtained shift amount. The method for estimating a spectral reflectance of a multi-band image according to claim 7. 9. The liquid crystal tunable filter including a liquid crystal unit, wherein the filter temperature correction is performed by comparing a temperature of the liquid crystal unit obtained from a thermometer in contact with the liquid crystal unit of the liquid crystal tunable filter. The relationship with the spectral transmittance distribution of the liquid crystal tunable filter is stored in advance as characteristic data, and the temperature of the liquid crystal unit at the time of capturing a multi-band image is
8. The spectral estimation of the spectral reflectance of a multiband image according to claim 7, wherein a shift amount of a wavelength of a spectral transmittance distribution of the liquid crystal tunable filter is obtained, and the spectral reflectance distribution of the subject is corrected according to the shift amount. Method. 10. Before performing the deconvolution processing, the number of spectral reflectance spectrum data is increased by nonlinear interpolation between the center wavelengths of the respective channels with respect to the spectral reflectance spectrum of the subject; The method for estimating a spectral reflectance spectrum of a multiband image according to claim 3, wherein the deconvolution process is performed using the spectrum of the increased spectral reflectance. 11. The spectral reflectance spectrum having an increased number of data has a wavelength interval of 10 nm or less.
6. The method for estimating the spectral reflectance of a multiband image according to the above. 12. A photographing means for photographing a subject, and when photographing the subject with said photographing means, a photographing wavelength region is divided into a plurality of channels, and a wavelength which transmits reflected light of the subject in a wavelength region corresponding to the channel. A variable filter and, for each channel, a conversion table in which a QL value obtained by photographing a chart having a known spectral reflectance is associated with the spectral reflectance in advance, and using the photographing means and the wavelength variable filter, A spectral reflectance spectrum estimating apparatus for estimating a spectral reflectance spectrum of a subject by converting a QL value of an original image of each channel of a multi-band image captured by using the conversion table into a spectral reflectance. A spectrum estimation system for a multi-band image, comprising: 13. The spectral reflectance spectrum estimating apparatus according to claim 13,
One device determined by a product of a spectral transmittance distribution of the wavelength tunable filter that transmits reflected light of a subject in a wavelength region of the channel, a spectral sensitivity distribution of an imaging unit, and a spectral intensity distribution of imaging illumination light during imaging. 13. The system for estimating the spectrum of a multiband image according to claim 12, wherein a deconvolution process is performed on the estimated spectral reflectance spectrum of the subject using a function to calibrate the spectral reflectance spectrum of the subject.